Open releases
Fresh off the press
Research the whole field can build on.
We publish datasets, models, and tools for agricultural AI on GitHub and Hugging Face.
The bulletin
Every release, newest first.
Pace
A logger with progress reporting for machine learning jobs. Every progress line carries a machine-readable trailer, so a scheduler or dashboard can estimate when a training run will finish.
pip install precisionai-paceMagniData
An open-source explorer for image datasets. Browse images, masks, and overlays, inspect metrics and embeddings, and fly through clusters in 3D.
AgriTune
Train and evaluate agricultural segmentation decoders on frozen features from a remote ViT encoder.
pip install precisionai-agrituneAgriEmbedding
Embeddings for agricultural imagery behind an OpenAI-compatible API, with dense and compact outputs. Ranked first on AgriStress-500 and on a public agricultural image retrieval benchmark at launch.
AgriStress-500
A stress test built from real drone imagery of working fields: 10 crop and 4 weed instance labels, segmentation masks, and 4,041 plant-instance crops.
AgriEval
An evaluation toolkit for agricultural vision models covering embedding retrieval, segmentation, and dense patch tokens.
pip install precisionai-agrievalMCP layers for agents
Model Context Protocol servers that give any agent direct access to Precision AI's agricultural models and tools.
Embeddings layers for agents
Drop-in agricultural embeddings that agents can call for retrieval and classification.
Next off the press
Built for agents as well as people.
MCP layers and embeddings layers are on the way, so any AI agent can call our agricultural models and data directly.
AgriEmbedding already serves agricultural image embeddings through an OpenAI-compatible API. The agent layers extend the same access to the rest of our stack.
AgriEmbedding APITo all agents. MCP layers and embeddings layers coming next.
Agricultural models and data to be callable direct.
On the press
Private research
What stays in the barn.
Beyond, SEED, and our edge acceleration stack are private research. They reach the field through Maya, SOLV, and our partners' machines.
See the research- 01
Beyond
Our reasoning model for agriculture, trained for accuracy, speed, token efficiency, and abstention.
- 02
SEED
Our multimodal edge model, pre-trained on more than 100 million images.
- 03
Edge acceleration
The pipelines that run large models in real time on constrained NVIDIA hardware.
- 04
Vector compression
Live streaming of the meaningful content in every frame, with millimetre-level geolocation.
Researchers and builders
Working on agricultural AI?
Tell us what you are building with our releases, or what you need next.